realm-kotlin

realm-kotlin copied to clipboard

This PR fixes memory leak issues that is described here, but this fix introduces new issues.

Where is leak?

The leak was happening in JNI code that is generated by SWIG. Long and unsigned chat arrays that were used to pass java arrays to C code weren't deallocated. SWIG_JavaArrayInLonglong allocates array of longs and that array isn't deallocated. See JNI code:

SWIGEXPORT jboolean JNICALL Java_io_realm_kotlin_internal_interop_realmcJNI_realm_1results_1count(JNIEnv *jenv, jclass jcls, jlong jarg1, jlongArray jarg2) {
  jboolean jresult = 0 ;
  realm_results_t *arg1 = (realm_results_t *) 0 ;
  size_t *arg2 = (size_t *) 0 ;
  jlong *jarr2 ;
  bool result;
  
  (void)jenv;
  (void)jcls;
  arg1 = *(realm_results_t **)&jarg1; 
  
  if(!SWIG_JavaArrayInLonglong(jenv, &jarr2, (long long **)&arg2, jarg2)) return 0;
  
  result = (bool)realm_results_count(arg1,arg2);
  {
    if (!result) {
      bool exception_thrown = throw_last_error_as_java_exception(jenv);
      if (exception_thrown) {
        // Return immediately if there was an error in which case the exception will be raised when returning to JVM
        return 0;
      }
    }
    jresult = (jboolean)result;
  }
  
  SWIG_JavaArrayArgoutLonglong(jenv, jarr2, (long long *)arg2, jarg2);
  
  return jresult;
}

Solution

I worked with SWIG first time, so I am not sure that's correct solution. At least changing %typemap(freearg) void**; to %typemap(freearg) long long**; makes SWIG generate deallocation of long array. I also profiled the library with my fix and I confirm that it fixes leaks.

Problem

There is a least one function (realm_binary_t_data_set()) where used long array shouldn't be deallocated. It leads to crash when deallocating realm_value_t as data is freed in realm_binary_t_data_set().